Development of a predictive model for benthic macroinvertebrates by using environmental variables for the biological assessment of Korean streams

摘要

Predictive models for benthic macroinvertebrates based on changes in environmental variables can assess the biological integrity of streams by comparing observed biotic communities with those expected at reference sites. To develop a predictive model of the abiotic community, we used benthic macroinvertebrates and environmental variables collected from 2,700 sites from 2010 to 2019. First, we selected 357 reference sites by using the 5-day biochemical oxygen demand, ammonia nitrogen, total phosphorus, turbidity, and coarse particle percentage. Then, we used Two-Way Indicator Species Analysis to classify the reference sites into six groups based on benthic macroinvertebrates. Reference sites classified by biological characteristics were linked to environmental variables by multi-discriminant analysis. The relative influences environmental variables on the classified groups were in decreasing order of catchment area, latitude, velocity, water depth, altitude, and longitude. To develop the predictive model, we combined (1) identification level, (2) grouping method, and (3) probability of capture, and then used the normalized root mean square error (NRMSE) to check the fit of each model. The higher the probability of capture was at the family level compared to the species level, the lower was the NRMSE. The grouping method was not as consistent as the identification level and probability of capture because the NRMSE for the number of taxa was low when used as a weighted average. The NRMSE was also low for the Benthic Macroinvertebrates Index and the Benthic Macroinvertebrates Family-level Biotic Index (BMFI) when used for assignment to the group with the highest probability. We selected the predictive model which used family level, weighted average, and BMFI-proposed indicator taxa as the final assessment model due to its sensitivity and fit. This model was the most reasonable choice, but we had to reduce the error of the model and revise it elaborately by securing additional environmental variables.